Pick models on the left, write one prompt, and run them in parallel. Costs are estimated before you generate.